print "nGuesses :", nGuess if __name__ == "__main__": imageReader = ImageReader() nHiddenLayer = 3 hiddenLayerSize = 1000 lmbda = 0.01 maxIter = 400 hiddenLayerSizes = [] for i in range(0, nHiddenLayer): hiddenLayerSizes.append(hiddenLayerSize) layerSizes = [imageReader.n] + hiddenLayerSizes + [imageReader.nGesture] (X, y), (Xv, yv) = imageReader.readImages() print len(X), "training data" print len(Xv), "test data" model = trainModel(layerSizes, X, y, lmbda, maxIter) print "nHiddenLayer :", nHiddenLayer print "hiddenLayerSize :", hiddenLayerSize print "lmbda :", lmbda print "maxIter :", maxIter print "Training", testModel(model, X, y, imageReader.nGesture) print "Test", testModel(model, Xv, yv, imageReader.nGesture)
import time import numpy as np from imageReader import ImageReader from keras.preprocessing.image import ImageDataGenerator from cnn import CNN if __name__ == "__main__": imageReader = ImageReader() cnn = CNN(imageReader.nGesture) (X, y), (Xv, yv) = imageReader.readImages(make1d=False, validationRatio=0.08) print len(X), "training data" print len(Xv), "test data" imageGenerator = ImageDataGenerator() cnn.train_gen(imageGenerator.flow(X, y, batch_size=16), len(X), Xv, yv) trainAccuracy = cnn.test(X, y) print "Training Accuracy :", trainAccuracy, "%" valAccuracy = cnn.test(Xv, yv) print "Test Accuracy :", valAccuracy, "%" #imageGenerator = ImageDataGenerator() #data = imageGenerator.flow_from_directory(directory='cropp', target_size=(100,100), color_mode='grayscale') #test = imageGenerator.flow_from_directory(directory='ASLval', target_size=(100,100), color_mode='grayscale') #cnn.train_gen(data, test) #trainAccuracy = cnn.test_gen(data, 11296)